Data-efficient Machine Learning for Engineering

In fields such as engineering, manufacturing, or healthcare, (labeled) data can be scarce and expensive. Unlike big data, the few available data are often of much higher quality. From a limited amount of data, knowledge must be created using various concepts such as active learning, surrogate modeling, transfer learning, self-supervised learning, few-shot learning, one-shot learning, synthetic data generation, hand-coded knowledge, and human-in-the-loop.

Using these data-efficient machine learning concepts, engineers can build and deploy effective models trained on small-scale datasets. Data-efficient approaches unlock numerous use cases where only small (labeled) datasets are available. This minimizes the time, engineering effort, and data required to create practical value with artificial intelligence.


A typical use-case for data-efficient Machine Learning (DEML) is design optimization based on computationally expensive physics-based simulations (e.g., finite elements methods (FEM) for electromagnetics or computational fluid dynamics). E.g., for a microstrip filter, the goal is to optimize its topology and electrical properties.

Research topics:

Machine Learning for Engineering (ML4ENG)

enabling technologies:

  • SUMO: SUrrogate MOdeling
  • DE-ML: Data-Efficient Machine Learning
  • AutoML: Automation in Machine Learning
  • PI-ML: Physics-Informed Machine Learning
  • X-ML: eXplainable Machine Learning


  • DT: Digital Twins (and siblings)
  • Design space exploration
  • Design and process optimization
  • Generative model-based design
  • UQ: Uncertainty Quantification